Unlocking the Future of Soil Health: Blending Data-Driven and Process-Based Models to Predict N₂O Emissions Across Sub-Saharan African Soils

Inorganic fertilisation is expected to increase across sub-Saharan Africa as a means to improve agricultural productivity, yet this expansion may substantially elevate nitrous oxide emissions, a potent greenhouse gas with high climate forcing. Nitrous oxide fluxes are characterized by strong spatial and temporal variability, often driven by short lived emission hotspots that are difficult to capture using conventional approaches. Identifying the dominant environmental and management controls on these emissions is therefore essential for robust prediction and mitigation. Despite this importance, sub-Saharan Africa remains one of the least studied regions, with fragmented observations and limited long term datasets. To address these challenges, we combine data driven machine learning approaches with process based ecosystem models to assess the drivers of nitrous oxide emissions across forest, grassland, and cropland systems. Using a harmonized dataset compiled from multiple sites and fertiliser treatments, we quantify annual nitrous oxide emissions and emission factors, and explore future emission trajectories under contrasting environmental conditions. Process based models are systematically tested, calibrated, and validated under sub-Saharan African soil and climate conditions to evaluate their performance and transferability beyond temperate regions where most models are traditionally developed. Our analysis highlights that data driven models show strong predictive skill, particularly for capturing nonlinear responses and emission hotspots, but their application is constrained by data scarcity and uneven spatial coverage. Process based models differ substantially in their data requirements and process representation. Simpler models such as DNDC and APSIM offer advantages for data limited regions, while models such as DayCent and STICS are better suited for hotspot and leaching focused analyses. Models resolving daily dynamics, including CNmodel and DAISY, provide valuable insights into seasonal variability but require detailed meteorological and deposition inputs. Overall, this study provides a structured evaluation of modelling strategies for nitrous oxide emissions in sub-Saharan Africa and identifies pathways to improve regional assessments. By aligning model complexity with data availability, our results support the development of climate smart, region specific mitigation strategies for nitrogen management in African agroecosystems.